How accurate are Heuritech’s forecasts?
On average, our forecasts have a median error of less than 10% at 12 months, across all geographies.
To measure accuracy, we use a standard backtesting method:
For a given season, we temporarily hide the last 12 months of data (i.e., the previous 4 seasons).
We run our forecasting models to predict how the trend would evolve during that hidden period.
We compare those predictions with the actual trend performance.
We calculate the difference using the ratio:
| actual - predicted | / actual,
based on the average trend volume for each season.
This process helps us continuously track and validate the performance of our models.